[tlinux-users:08133] Re: New release of omnibook module soon: testing welcome

Holger Nelson a écrit :

> I get a null-pointer-dereference in nbsmi_smi_command on my Toshiba Tecra
> S1 and narrowed it down to these lines:
>                         save_all_regs_flag();
>                         intel_do_smi_call(&retval,function,sci_en);
>                         restore_all_regs_flag();
>
> I will try to debug this after I slept. 
Well I have two theories:
-There is an obvious bug that I don't see (quite possible)
-The spec I have say that writing to port b2 or b0 trigger
a System Management Interrupt which may mangle all
the registers* an flags of  the processor , hence the
'pusha pushf' and 'popa popf' which save an then restore
them.
"save_all_regs_flag();
intel_do_smi_call(&retval,function,sci_en);
restore_all_regs_flag(); "

Generate this code for me:
  d8:    60                       pusha 
  d9:    9c                       pushf 
  da:    8d 44 24 0c              lea    0xc(%esp),%eax
  de:    8b 54 24 10              mov    0x10(%esp),%edx
  e2:    89 44 24 08              mov    %eax,0x8(%esp)
  e6:    ed                       in     (%dx),%eax
  e7:    89 c1                    mov    %eax,%ecx
  e9:    31 c0                    xor    %eax,%eax
  eb:    ef                       out    %eax,(%dx)
  ec:    89 f0                    mov    %esi,%eax
  ee:    66 e7 b2                 out    %ax,$0xb2 <--- Trigger SMI
  f1:    89 44 24 08              mov    %eax,0x8(%esp)
  f5:    89 c8                    mov    %ecx,%eax
  f7:    ef                       out    %eax,(%dx) <--- If dx has been 
cleared we are doomed
  f8:    9d                       popf  
  f9:    61                       popa

Which mean that I will have to recode the
whole block in asm to control precisely what is going on.
This might also explain the failures reported by Renato S. Yamane
 
* While not precised anywhere in the doc I have assumed it does
not include esp or I fail to see how we can recover.

-- 
Mathieu